English construction “be easy to VERB”

[dan-zeman]

How should we annotate sentences like This is easy to clean in English? The current (UD 2.11) annotation in various English treebanks (EWT, GUM, LinES, ParTUT, PUD, Pronouns) is inconsistent.

• It is easy to make your contribution. EWT: expl(easy, It); cop(easy, is); csubj(easy, make); mark(make, to); obj(make, contribution). This approach is also found in GUM.
• It is not an easy matter to assess how the savings affects our standard of living. ParTUT: expl(matter, It); cop(matter, is); amod(matter, easy); csubj(matter, assess); mark(assess, to); ccomp(assess, affects)
• It's not easy to separate fact from fiction. EWT: nsubj(easy, It); cop(easy, 's); ccomp(easy, separate); mark(separate, to); obj(separate, fact); obl(separate, fiction)
• It is easier to run a hotel. GUM: expl(easier, It); xcomp(easier, run); mark(run, to); obj(run, hotel)
• Hers is easy to clean. Pronouns: nsubj(easy, Hers); cop(easy, is); csubj(easy, clean); mark(clean, to)
• No pet is easy to take care of. EWT: nsubj(easy, pet); cop(easy, is); ccomp(easy, take); mark(take, to); obj(take, care); obl(take, of)
• The effect was easy to measure. GUM: nsubj(easy, effect); cop(easy, was); xcomp(easy, measure); mark(measure, to). This approach is also found in PUD.
• The staff were easy to get along with. EWT: nsubj(easy, staff); cop(easy, were); advcl(easy, get); mark(get, to); advmod(get, along); obl(get, with). This approach is also found in ParTUT.
• Jack Dempseys are not an easy cichlid to breed. EWT: nsubj(cichlid, Dempseys); cop(cichlid, are); amod(cichlid, easy); advcl(easy, breed); mark(breed, to)
• That's the easy one to provide for. EWT: nsubj(one, That); cop(one, 's); amod(one, easy); acl(easy, provide); mark(provide, to); obl(provide, for). This approach is also found in GUM and LinES.
• The concept is an easy one to dismiss. GUM: nsubj(one, concept); cop(one, is); amod(one, easy); acl(one, dismiss); mark(dismiss, to)

Obviously, the approach of the Pronouns treebank to Hers is easy to clean is an error because there are two subjects under the same predicate. (This is actually how I ran across this issue.) However, it also highlights the main disunity in the observed approaches: is the to clause a subject or not? The other treebanks make the to clause subject (with a few exceptions) if the pre-verbal argument is the pronoun it and the pronoun can be understood as non-referential. Otherwise they make the preverbal nominal the subject and the to clause is something else. Note however that it would be possible (although none of the treebanks does it) to always make the to clause subject. Then the preverbal argument would be a (non-projective) object: csubj(easy, clean); obj(clean, Hers).

When the to clause is not a subject, the observed relations are ccomp, xcomp, advcl and acl; that is, all possible clausal dependents. Of these, acl is clearly wrong if the head is the adjective easy (and not a noun modified by easy). The xcomp relation looks appealing at the first sight, because there is a coreference between arguments (which could be made explicit in the enhanced representation) but it is wrong because here the shared argument is not the subject of the to clause. For xcomps, the shared argument may be subject or object or oblique dependent of the matrix verb, but in the xcomp clause it always corresponds to the subject. That leaves us with ccomp and advcl. I lean towards advcl because I can't really see easy (or any other adjective) as a transitive predicate.

If easy is an amod of a noun, should the to clause be attached to easy (non-projectively), or to the noun? Attaching it to easy is attractive (despite the non-projectivity) because it does not change attachment when the noun is not present. However, some nouns may themselves take a to clause (cf. it is an easy way to escape vs. it is a way to escape).

Also, if it is easy for someone to do, then there is another question what to do with for someone. Some treebanks make it subject of the subordinate verb, some (ParTUT) make it oblique dependent of easy.

[nschneid]

Related: #308

[nschneid]

^ This is a long thread and I don't have the bandwidth to process it now, but my gut feeling at present would be:

• Treat the "it" sentences as extraposition with the infinitival clause as csubj.
• If the adjective licensing the infinitival clause is an amod in a nominal, attach the infinitival clause to the nominal head. CGEL calls this relationship an "indirect complement". (Otherwise, it would be awkward to treat It is not an easy matter to assess as extraposition. The non-extraposed counterpart is To assess is not an easy matter, which is clearly csubj(matter,assess), not csubj(easy,assess).)
• Hers is easier to clean: As Dan suggests, I think we should consider csubj(easier,clean), obj(clean,hers), since this paraphrases as To clean hers is easier. Sort of a topicalized extraposition?
  • Downside: The infinitival can be dropped with enough context (ellipsis?): Mine is hard to clean, but hers is easier. So the two analyses of the copular construction would look artificially different.
  • An alternative is to use the new advcl:relcl relation, like for it-clefts (It's John I want to work for). nsubj(easier, hers), advcl:relcl(easier, clean). Then we have an Enhanced edge E:obj(clean, hers). I wouldn't claim this is EXACTLY a relative clause, but it has some similar properties to relative clauses, so advcl:relcl might be good enough for our purposes.
• The concept is an easy one to dismiss: This is clearly not extraposition (*To dismiss this concept is an easy one). Reminds me of an infinitival relative, e.g. a good bed to sleep in, or a good bed to buy. Maybe acl:relcl(one, dismiss). This allows us to capture in Enhanced Dependencies that "dismiss" is not missing an object.

[AngledLuffa]

Hers is easier to clean vs To clean hers is easier: This is also quite similar to Cleaning hers is easier, I would think. That can also be dropped in a longer context: Cleaning mine is hard, but hers is easier

[nschneid]

Hers is easier to clean vs To clean hers is easier: This is also quite similar to Cleaning hers is easier, I would think. That can also be dropped in a longer context: Cleaning mine is hard, but hers is easier

Hmm, so you're saying that "Mine is hard to clean, but hers is easier" can be interpreted as "Mine is hard to clean, but cleaning hers is easier". If that were the analysis, shouldn't we treat "hers" as standing for a csubj?

Another thing to consider: standard tests of subjecthood, like case and agreement—"They are easier to clean" vs. "It is easier to clean them"; "That is easier to clean" vs. "Those are easier to clean". Whereas we can say expletives are special, I find it harder to justify treating they/that/those as non-subjects.

[Stormur]

The other treebanks make the to clause subject (with a few exceptions) if the pre-verbal argument is the pronoun it and the pronoun can be understood as non-referential. Otherwise they make the preverbal nominal the subject and the to clause is something else.

I am with the observations by @nschneid here that this is the best analysis. It is quite clearly defined, in that when the subject is clausal, and so "heavier", it seems to tend to be shifted away and replaced by the expletive it. But a full nominal subject stays there as usual.

The xcomp relation looks appealing at the first sight, because there is a coreference between arguments (which could be made explicit in the enhanced representation) but it is wrong because here the shared argument is not the subject of the to clause.

I would stay with xcomp also because I do not see alternatives: advcl is not justified, it would be too loose.

I would put forth as a consideratio nthe actual indeterminacy of "non finite" verb forms with respect to Voice: it seems to be the case in English that the infinitive form is ambivalent in some cases. I would try to make a confront with Italian, where I have to change the infinitival clause according to voice:

• eng. It is easy to clean [the table] > È facile pulirla ('is easy clean-her')
  • where you have la referring to tavola 'table' (feminine) , and pulire is the infinitive (it has only one form): so this clause is transitive and active
• eng. The table is easy to clean > La tavola è facile da pulire ('The table is easy from clean')
  • such an infinitive introduced by the ADP da 'from' is passive: you cannot say da pulirla (at least not with the same meaning).
  • if you say è facile da pulire the sentence is not impersonal like the first one, but an actual subject is implied, the same of da pulire.

Theoretically you might say something like facile da essere pulita, but id does not sound so well, or, at least, it is plain pleonastic. Probably similar considerations are valid for easy to be cleaned?

The concluding observation is that there is no particular value of Voice to attach to an infinitive form. It might take part mostly in active constructions, but it doubles in passive ones, too. There can be ways to make it explicit (like in Italian), but probably English has no need to do so.

If easy is an amod of a noun, should the to clause be attached to easy (non-projectively), or to the noun?

(cf. it is an easy way to escape vs. it is a way to escape)

To the noun: I think we are speaking of different things here, it is not the adjective licensing that infinitival clause, but they just rely on the same strategy.

As said by @nschneid , I also agree here that in both cases the head is the noun and the verbal element is its acl.

Also, if it is easy for someone to do, then there is another question what to do with for someone. Some treebanks make it subject of the subordinate verb, some (ParTUT) make it oblique dependent of easy.

I agree with the clausal interpretation, but observe that in this case for is the marking ADP of someone to do, with a non finite predicate. So:

mark(do,for)
nsubj(do,someone)

Probably in this interpretation we have an advcl(easy,do), since it is the equivalent of a nominal obl as in easy for beginners.

[nschneid]

Returning to this issue after awhile. To boil it down with some simpler examples, I think we are dealing with the following (setting aside "It is easy to clean.", which is ambiguous):

1. To clean/Cleaning(csubj) this shirt is easy(root).
2. It(expl) is easy(root) to clean(csubj) this shirt.
3. This shirt is easy(root) to clean.
4. This shirt is a pain(root) to clean.
5. This is an easy shirt(root) to clean.
6. This is an easy-to-clean shirt(root).

(1) and (2) seem straightforward. I have included the deprels inline.

The rest look like copular predications, so I am assuming the root is the head of the predicate phrase.

(6) looks like an attributive compound so not necessarily analyzed with clause-level structure.

The main question then is where and how to attach "to clean" in (3-5). Specifically: (i) Is it a clause-level constructional dependent (cf. advcl:relcl for it-clefts)—suggesting advcl—or is it licensed by the "tough" item (ccomp)? (ii) If licensed by the "tough" item, does that mean it is adnominal (acl) in (4)?

Note that this is an construction with an extracted argument (This shirt_i is easy to clean ___i). But it is not xcomp: xcomp means the subordinate clause subject is inherited from an argument position of the matrix clause. In this example it is the object that is gapped.

From @Stormur's comments on Italian above and from what @mcdm has said about French, I suspect the Romance tough-constructions may be different in important ways from English. If the embedded verb is passive ('The table is easy of being-washed'), maybe xcomp is indeed appropriate.

[dan-zeman]

I think that to clean is licensed by easy in (3-5). You can replace easy by a few other adjectives that can license an infinitive complement (difficult, tough) but you cannot put any adjective there (?The shirt is green to clean.) Also, omitting the infinitive complement feels like ellipsis (?The shirt is easy. vs. The shirt is green.) So it looks somewhat like ccomp.

With other lexical material, you could have a nonverbal predicate modified by an infinitival adverbial clause: The shirt is here/prepared/ready to be cleaned. I'm not sure how easy it will be to distinguish these cases from the licensed ones above. (Maybe the infinitive has to be passive in advcls because the subject of an adverbial infinitive must be coreferential with the subject of the matrix clause?) It still bugs me that such "ccomps" cannot be substituted by a nominal direct object, which is what we normally expect. So I'm not particularly happy with ccomp for the former case in the UD framework (where licensing and obligatoriness are not considered important), and I would be probably fine with using advcl in both situations, although it does not feel perfect for easy to clean.

If we acknowledge that the infinitive is licensed by the "tough" item and that it matters for the UD analysis, then the infinitive must be acl in (4) because it modifies the noun and not the clause (if it modified the clause, it could be advcl despite being attached to a noun). And in (5) the ccomp must be attached nonprojectively to easy (while an advcl could be attached to shirt (because it modifies a clause) or to easy (becuase it modifies an adjective)).

[nschneid]

(Maybe the infinitive has to be passive in advcls because the subject of an adverbial infinitive must be coreferential with the subject of the matrix clause?)

Some of these may be controlled-subject clauses (whether xcomp or adjunct). But "The chicken is ready to eat" is ambiguous:

N.B. What CGEL terms "hollow" clauses are a superset of the tough-construction clauses; hollow clauses also include those licensed by such items as "ready", "good", "worth", "take" with a duration, and the sufficiency/excess degree markers "too", "enough", etc. (SIEG2 pp. 320-321).

It still bugs me that such "ccomps" cannot be substituted by a nominal direct object, which is what we normally expect.

There are some verbs, like wonder, that license ccomp but not obj. But yeah, it would strengthen the case for ccomp if tough-constructions alternated with obj at least some of the time. It may just be a consequence of ccomp being prototypically for complements of verbs, as opposed to adjectives, which usually don't take objects.